Limited biomass recovery from gold mining in Amazonian forests

1. Gold mining has rapidly increased across the Amazon Basin in recent years, es pecially in the Guiana shield, where it is responsible for >90% of total deforesta tion. However, the ability of forests to recover from gold mining activities remains largely unquantified.


| INTRODUC TI ON
Within the last decade, gold mining activity has become more prevalent across the Amazon Basin, to the extent that it is now the principal driver of deforestation in several northern Amazonian countries and a significant factor elsewhere (Alvarez-Berríos & Mitchell Aide, 2015;Asner, Llactayo, Tupayachi, & Luna, 2013;Guyana Forestry Commission & Indufor, 2012). In Guyana, for example, gold mining led to ~57,000 ha of forest lost between 2010 and 2017 (Guyana Forestry Commission, 2018), representing an area of similar magnitude to that in Peru, where gold mining-related deforestation was estimated at >60,000 ha within the same timeline (Espejo, Messinger, Román-Dañobeytia, Ascorra, & Luis, 2018). Much of this forest loss stems from artisanal and small-scale miners, who respond rapidly to increases in international gold prices (Howard et al., 2011). These mining activities invariably lead to widespread environmental damage and deforestation, resulting in substantial soil erosion and contamination, increased forest fragmentation and mercury pollution of rivers and streams (Castilhos et al., 2015;Dedieu, Allard, Vigouroux, Brosse, & Céréghino, 2014;Sonter et al., 2017;Veiga, Maxson, & Hylander, 2006). Mercury bioaccumulates along the aquatic food web and the consumption of contaminated fish, an integral part of human diet, can result in fatal damage to humans' central nervous system in affected regions (Gutiérrez-Mosquera et al., 2018). Further ecological impacts may arise from hunting and/or land clearance to support mining settlements (Sonter, Ali, & Watson, 2018).
The extent of soil physical damage and chemical contamination associated with gold mining activities sets it apart from other traditional deforestation drivers such as conversion to pasture or small-scale agriculture, which generally do not significantly alter soil structure or nutrients (e.g. nitrogen and phosphorus; Santos-Francés, García-Sánchez, Alonso-Rojo, Contreras, & Adams, 2011;Wantzen & Mol, 2013). The excavation of topsoil associated with mining renders the soil deficient in organic matter and nutrient content (Roman-Danobeytia et al., 2015). As such, gold mining is expected to arrest forest recovery rates compared to other land uses. Additionally, the toxicity from mercury applied during mining may be harmful to microbes, adversely affecting plant physiology and metabolism once threshold soil concentrations are breached (Beauford, Barber, & Barringer, 1977).
Despite the importance of mining as a major driver of tropical deforestation, its impacts on forest biomass accumulation remain largely unquantified. We are aware of only one previous field study that has attempted to evaluate forest recovery following deforestation from gold mining in the Amazon (Peterson & Heemskerk, 2001), but this study performed only visual assessments of vegetated area following 1-4 years of recovery with no quantitative measurements. With the increasing availability of high-resolution satellite imagery, other authors such as Novoa, Finer, and Román (2016) also visually assessed forest regeneration following gold mining, finding vegetation regrowth in previously mined areas in Peru. However, such studies have not incorporated field-based ground-truthing and do not consider separately the complex nature of mining sites, which typically consist of three distinct mining zones: (a) overburden: areas overlying the gold ore, including the topsoil, which are displaced during the mining process, (b) tailing pond: deposits of material remaining after the gold has been separated from the ore and (c) the mining pit.
The intensity of the disturbance caused by mining is much higher in the tailing pond and mining pit. In these mining zones, recovery can be viewed as primary succession due to the negligible biological legacy that remains in the soil, resulting in a new surface for colonization that does not resemble the pre-disturbance conditions (Prach & Walker, 2019). Conditions for recovery are much more favourable in the overburden, however. Here, recovery is viewed as a secondary succession process, due to greater biological legacy remaining in the soil following the disturbance.
Indeed, across the tropics, secondary forests are thought to constitute a significant carbon sink, which in some estimates are even larger than that of intact primary forests (Pan et al., 2011). Recent syntheses of chronosequences from tropical secondary forests have concluded that secondary forest biomass recovery rates are (a) largely controlled by background rainfall conditions  and (b) faster in forests previously under pasture than 7. Synthesis and applications. To achieve large-scale restoration targets, Guyana and other Amazonian countries will be challenged to remediate previously mined lands.
The recovery process is highly dependent on nitrogen availability rather than mercury contamination, affecting woody biomass regrowth. The significant recovery in overburden zones indicates that one potential active remediation strategy to promote biomass recovery may be to backfill mining pits and ponds with excavated soil.

K E Y W O R D S
Amazon, biomass recovery, forest recovery, forest restoration, gold mining, land use change, secondary forests, tropical forest agricultural production (Martin, Newton, & Bullock, 2013). Yet none of these syntheses contain chronosequences on previously mined areas, and thus we do not know how recovery of biomass and biodiversity following gold mining compares with recovery from more traditional land uses such as pasture and agriculture.
To address these critical knowledge gaps, we established forest plots in focal gold mining regions in Guyana to investigate the patterns of biomass recovery following the abandonment of gold mining activity, re-censusing them 18 months later. Here, we specifically consider (a) biomass accumulation between 3 and 4 years in recently abandoned gold mining plots, (b) how observed biomass recovery patterns from gold mining at our study sites compare with published tropical forest chronosequence data from other land uses (e.g. pasture, agriculture) and (c) which factors most limit biomass recovery in previously mined sites. Finally, having provided the first detailed field-based information on the regeneration of secondary forests after gold mining activity, we explore the implications of our findings for gold mining recovery in the wider Amazon context, and provide the first ground-based estimate of the secondary forest carbon sink foregone as a result of gold mining activities. All tree saplings 25-200 cm tall and all tree seedlings 5-25 cm tall were identified and counted in the 3 m × 3 m and 1 m × 1 m subplots respectively. Nested subplots were also established in the 1-ha control plot, where all trees >10 cm DBH were measured and recorded and standard RAINFOR protocols applied to map, measure and identify trees (Phillips, Baker, Feldpausch, & Brienen, 2015). Tree species in all plots were identified by a local botanist.

| Study sites and sampling design
Heights of larger trees, saplings and seedlings were also measured (see Appendix S1 for details).

| Biomass calculations
To calculate above-ground biomass (AGB) of individual trees (in kg), we used the allometric model of Chave et al. (2014): where ρ is the wood density (g/cm 3 ), D is the DBH (cm) and H is the height (m). This is a generic equation for tropical forests, based on a broad set of trees including primary and secondary forest species, which has previously been used for estimating biomass in young secondary forests Rozendaal et al., 2017). Wood density values for species were obtained from the global wood density database available from the Dryad data repository (http://datad ryad.org/).
Where species-specific wood densities were not available, genus or family mean wood densities were used (e.g. Baker et al., 2004). As the relative proportions of overburden, mining pit and tailing pond zones varied across plots, we rescaled our total biomass estimates for each zone so that all values are expressed on a 1-ha basis. For each plot, we calculated the dry AGB in Mg/ha, and the rate of net change in biomass between our two measurement censuses (∆AGB), the biomass increase due to diameter growth (∆AGB Growth ), the biomass increase due to recruitment of new trees (∆AGB Recruitment ) and the biomass loss due to mortality (∆AGB Mortality ), all in Mg/ha per year.

| Comparison with published secondary forest chronosequences
To place our study within the wider literature on tropical forest biomass recovery, we compared our AGB and ∆AGB values with published data for other tropical forests at 2-4 years since abandonment from chronosequence studies of recovery from prior pasture and agriculture land use. Specific criteria for inclusion of secondary forest biomass recovery studies in this analysis are provided in Appendix S1.

| Collections and analysis of soil samples
Soil samples were collected between January and March 2016 at the time of plot installation and determined for cation exchange capacity (CEC), total nitrogen (TN) content, total phosphorus (TP) content, mercury concentration and particle size distribution, following well-established protocols (see Appendix S1 for details of soil analysis).

| Statistical analysis
To examine the effects of mining regions and individual mining zones on AGB, ∆AGB, AGB Mortality , AGB Recruitment and AGB Growth , linear mixed-effects models were used, with a random intercept for each plot. For information on statistical treatment of biomass changes across plots, soil data and comparison with published estimates from other studies please refer Appendix S1.

| Above-ground biomass dynamics during secondary succession
Although standing ABG biomass estimated in the control, old-growth plot in Mahdia (607 Mg/ha) was considerably higher than in the control plot in Puruni (409 Mg/ha), biomass accumulation following mining was markedly lower in Mahdia than in Puruni (Figure 3). In Puruni, Zones of a small-scale mining process: (i) overburden: areas overlying the ore which are displaced during the mining process, (ii) tailing pond: deposits of material left over after the gold has been separated from the ore, and the (iii) mining pit. Size of entire plot is 0.25 ha. Area of zones differed from 0.05-0.18 ha for overburden, 0.03-0.13 ha for tailing pond and 0.03-0.16 ha for mining pit. Image © M. Kalamandeen (2016) overburden woody AGB following 3-4 years of forest recovery was over six times that in Mahdia (6.63 ± 1.63 Mg/ha vs. 1.06 ± 0.38 Mg/ha).
However, woody AGB was almost negligible in mining pits and tailing ponds (Figure 3). In fact, no woody species were observed in five of the nine mining pits and five of the nine tailing pond plots (i.e. woody AGB = 0) following 3-4 years of recovery and the maximum recorded woody AGB in Puruni and Mahdia was 0.39 and 0.29 Mg/ha respectively. Additionally, no trees >10 cm in height were found in any of our mining pit or tailing pond plots, even in the second census. In both sites, virtually all of the AGB in pits and ponds was accounted for by grasses and sedges, such as Andropogon bicornis. Based on destructive harvesting undertaken in Census II, herbaceous (non-woody) AGB was estimated at 0.03 ± 0.03 Mg/ha in Puruni and 0.17 ± 0.11 Mg/ha in Mahdia for mining pits and 0.09 ± 0.07 Mg/ha (Puruni) and 0.2 ± 0.08 Mg/ha (Mahdia) for tailing ponds.
Annual rates of overburden above-ground biomass change (ΔAGB) differed strongly between sites and zones (

| Forest composition and structure
Overall, there was a clear difference in floristic composition between control and overburden plots ( Figure S1). Fabaceae contributed most to AGB in the control plots, accounting for 63% of total AGB in both Mahdia and Puruni, followed by Lecythidaceae (16% and 26% of total AGB respectively). Differences in familial composi-

| Soil nutrient and mercury levels
To understand the differences in biomass and species composition, soil analyses were conducted, indicating significant differences across mining zones for mercury concentration (p < 0.0001), TN (p = 0.004) and exchangeable cations (p = 0.04; Table S3; Figure 5).
Mercury concentrations in overburden zones were on average 55% and 72% lower than tailing ponds and pits in Mahdia and Puruni respectively (Table S3; Figure 5). Although mining pits and ponds were only abandoned for 6 months to 3 years, their mercury concentrations were two orders of magnitude lower than active site concentrations ( Figure 5), suggesting that leaching of mercury from soils of mining areas happens very quickly. Further details on variation of TP, TN and soil bulk density across plots are provided in Appendix S2.

| Controls on biomass recovery
Among all predictors, TN had the strongest effect, featuring in all the best supported models for AGB, AGB Growth and AGB Mortality . It was

F I G U R E 4
The contribution of tree growth, recruitment and mortality compared to annual biomass change on the overburden zone between Census I and Census II for abandoned mining plots in Mahdia and Puruni F I G U R E 5 Mercury (Hg), total nitrogen (TN) and total phosphorus (TP) concentrations (mg/kg) in soils from mining zones (overburden, tailing pond, mining pit), in active mines and in oldgrowth forest control sites at Mahdia and Puruni also the single most important predictor of mortality and explained 58% of the variations in AGB Growth . Combinations of TN and TP, mercury and exchangeable cations had the strongest effect on AGB CensusII and ΔAGB (Table S6), with these models explaining 38%-43% of the observed variation in AGB. Overall predictive ability of the models was relatively weak for both AGB Mortality and AGB Recruitment where the best models explained only 20%-25% of the variation. Mercury content was found to be the most important predictor of AGB Recruitment , implying an important role in limiting tree establishment. Phosphorus content was found to be less important than other predictors in driving variation in biomass recovery across plots.  (Table S2), our results are qualitatively consistent with those of Peterson and Heemskerk (2001)

| Factors influencing biomass recovery
The lack of woody biomass recovery, particularly on the mining pit and tailing pond of gold mining sites, was found to be heavily F I G U R E 6 Recovery of biomass following different land use, relative to undisturbed reference forests F I G U R E 7 Comparison of annual biomass change of abandoned gold mining sites (overburden only) with three second-growth (5-15 years of age) wet forests across the Neotropics (MAN = Manaus, Brazil; CHJ = Chajul, Mexico; SAR = Sarapiqui, Costa Rica; Rozendaal et al., 2017) influenced by available TN and mercury concentrations (Table S6).
Although we did not find a strong role for phosphorus in driving biomass accumulation, fertilization experiments have shown that phosphorus additions to soil, when combined with nitrogen additions, can also boost biomass recovery in secondary forests Davidson, Reis de Carvalho, et al., 2004). In our study, the differences in phosphorus content between overburden plots, which recovered well, and plots on previously mined tailing ponds and mining pits were not as pronounced as for nitrogen. Previously mined tailing ponds and mining pits, which showed no recovery in both sites, were found to contain 64%-94% of the soil phosphorus content of the overburden plots.
While available nitrogen probably drives the differential recovery observed across mining zones, our analyses suggest that other factors likely drive the differences in accumulation rates observed between the two study sites. Indeed, overburden biomass accumulated approximately six times more slowly in Mahdia than in Puruni, despite actual nitrogen and phosphorus concentrations in overburden soils being on average higher in Mahdia than in Puruni (Table S4). Soil mercury content and exchangeable cations emerged as important secondary predictors of biomass accumulation metrics (Table S6) and may explain the observed between-site differences.
Mercury concentrations on overburden soils in Mahdia were ~3.5 times greater than those in Puruni and were higher than mercury concentrations in previously mined pits and tailing ponds in Puruni.
This may be due to higher mercury concentrations used in the mining process and/or the progressive frequency of mining activities occurring in Mahdia vs. Puruni.
Active mining sites were also substantially higher in mercury levels than the abandoned mining zones suggesting that the majority of the mercury is not retained locally in the soils and may be transported to surrounding soils and waterways. Other studies have shown that mercury-rich tailings have been detected hundreds of kilometres away from mining sites (Diringer et al., 2015). The potential rapid mercury leaching to neighbouring areas may lead to larger ecological impacts than previously estimated. Given the estimated level of mercury in Amazonian soils is between 0.02 and 0.4 mg/kg (Santos-Francés et al., 2011), mercury concentrations in active mines at Puruni and Mahdia were three to 240 times higher than reported levels respectively. Furthermore, only one overburden plot in Puruni was within the reported range of mercury concentrations seen elsewhere.
Our statistical analyses suggest an important role for mercury in inhibiting recruitment (Table S6). Mercury concentrations were previously found to strongly inhibit plant development and growth, but have differential impacts across taxa (Muddarisna & Krisnayanti, 2015).
Comparison of plant composition data for overburden plots at both sites revealed that an important difference between the compositions of the overburden vegetation is the near absence of leguminous woody species with the capacity to fix nitrogen in Mahdia (Figure 8).
While abundance of Fabaceae is similar in control plots in both sites, species capable of fixing nitrogen were significantly more abundant in Puruni. Nitrogen fixers are known to drive biomass accumulation in recovering secondary forests (Batterman et al., 2013) and their suppression in Mahdia may be associated with higher mercury loads inhibiting the activity of nitrogen-fixing bacteria (Nuraini, Arfarita, & Siswanto, 2015;Rafique, Amin, & Latif, 2015). Some studies on agricultural crops have shown that nitrogen-fixing bacteria from the genera Staphylococcus, Pseudomonas, Escherichia and Bacillus may be resistant to some levels of mercury contamination by converting mercury into a less toxic form, such as hydrogen sulphide (Nuraini et al., 2015;Rafique et al., 2015). We hypothesize that similar mechanisms may be limiting the establishment of nitrogen-fixing trees in Mahdia, yet we are aware of no study that has examined this. Future studies should directly test the hypothesis that germination of tropical nitrogen-fixing trees and association with nitrogen-fixing bacteria is reduced by soil mercury.

| Implications for policy and rehabilitation
Relative to other land uses, biomass recovery in previously mined pits and tailing ponds are among the lowest recorded in the literature (Figures 6 and 7). Certain species can also play an important role in phytoremediation within post-mining sites (Ritchie & Raina, 2016), either through bioaccumulation (i.e. survival despite concentrated contaminants) or through bioexclusion (i.e. by restricting contaminants uptake into their biomass; Tangahu et al., 2011). Although we did not formally test for the tolerance of specific species to mercury, we note that several tree taxa were found on the overburden plot in Mahdia, despite high soil mercury concentrations. These species, which include Vismia guianensis and Cecropia obtusa, may therefore represent candidate species for more focused phytoremediation testing. The optimal selection of tree taxa for restoration requires carefully planned field trials as the germination/establishment sensitivity of different taxa to high mercury concentrations remains largely unstudied.
Given their central role in driving biomass recovery, phytoremediation strategies which promote the functional integrity of nitrogen fixation processes need to be further studied. Ultimately, however, large-scale remediation efforts will hinge on a balance of costs and as an important element of its program for reducing carbon emissions. The slow recovery rates observed in our study suggest that active remediation approaches will be necessary for the success of such policies, along with the systematic monitoring of gold mining activities and enforcement to provide quantitatively robust evidence of change over time.